Research map / Architectures and efficiency / RNNs and normalization
An Intriguing Property of Geophysics Inversion
Yinan Feng, Yinpeng Chen, Shihang Feng, Peng Jin, Zicheng Liu, Youzuo Lin
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace 1D sine encoder kernel with sin(n pi t)cos(n pi x) (ablation) | InvLINT · Marmousi | MAE: 136.67 -> 128.33 | Improved |
| replace 1D sine encoder kernel with sin(n pi t)cos(n pi x) (ablation) | InvLINT · Kimberlina-Leakage | MAE: 8.13 -> 8.88 | Hurt |
| replace Gaussian decoder kernel with sin(n pi x)sin(n pi z) (ablation) | InvLINT · Marmousi | MAE: 136.67 -> 177.36 | Hurt |
| use Gaussian with 1/3 sigma or sinc kernel instead of Gaussian decoder kernel (ablation) | InvLINT · Marmousi | MAE: 136.67 -> 138.19 | Hurt |
| replace transformer+linear decoder with Conv x2 + Linear (ablation) | InvLINT · Marmousi | MAE: 136.67 -> 140.72 | Hurt |
| replace transformer+linear decoder with Deconv x1 + Linear (ablation) | InvLINT · Marmousi | MAE: 136.67 -> 167.98 | Hurt |
| use deeper (Deconv + Conv) x5 decoder instead of 1 transformer layer (ablation) | InvLINT · Marmousi | MAE: 136.67 -> 126.59 | Improved |
| use deeper (Deconv + Conv) x5 decoder instead of 1 transformer layer (ablation) | InvLINT · Kimberlina-Leakage | MAE: 8.13 -> 6.86 | Improved |
| use unshared multiple linear layers instead of a shared linear layer (ablation) | InvLINT · Marmousi | MAE: 136.67 -> 138.82 | No clear effect |
| increase transformer depth from 1 to 3 layers (ablation) | InvLINT · Marmousi | MAE: 136.67 -> 132.19 | Improved |
| increase transformer depth from 1 to 3 layers (ablation) | InvLINT · Kimberlina-Leakage | MAE: 8.13 -> 8.09 | No clear effect |
| reduce kernel number N from 2048 to 1024 (ablation) | InvLINT · Kimberlina-Leakage | MAE: 8.13 -> 8.63 | Hurt |
Closest papers by meaning
- Unsupervised Learning of Full-Waveform Inversion: Connecting CNN and Partial Differential Equation in a Loop (ICLR 2022 · similarity 0.90)
- A Unified Framework for Forward and Inverse Problems in Subsurface Imaging using Latent Space Translations (ICLR 2025 · similarity 0.89)
- Raw Nav-merge Seismic Data to Subsurface Properties with MLP based Multi-Modal Information Unscrambler (NeurIPS 2021 · similarity 0.87)
- Auto-Linear Phenomenon in Subsurface Imaging (ICML 2024 · similarity 0.87)
- Physics-Informed Deep Inverse Operator Networks for Solving PDE Inverse Problems (ICLR 2025 · similarity 0.81)
- Neural Field Thermal Tomography: A Differentiable Physics Framework for Non-Destructive Evaluation (NeurIPS 2026 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
